Re: [PATCH] blame: add the ability to ignore commits

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 15, 2019, 06:08 UTC
Message-ID: <xmqqd0oytpv2.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <003c01d4ac36$bdf29470$39d7bd50$@nexbridge.com>

"Randall S. Becker" <rsbecker@nexbridge.com> writes:

Show 31 quoted lines

> On January 14, 2019 12:46, Junio C Hamano wrote:
>> Barret Rhoden <brho@google.com> writes:
>> 
>> > On 2019-01-10 at 14:29 Junio C Hamano <gitster@pobox.com> wrote:
>> >> > For instance, commit X does this:
>> >> >
>> >> > -foo(x,y);
>> >> > +foo(x,y,z);
>> >> >
>> >> > Then commit Y comes along to reformat it:
>> >> >
>> >> > -foo(x,y,z);
>> >> > +foo(x, y, z);
>> >> >
>> >> > And the history / rev-list for the file looks like:
>> >> >
>> >> > ---O---A---X---B---C---D---Y---E---F
>> >> >
>> >> > I want to ignore/skip Y and see X in the blame output.
>> >>
>> >> If you skip Y, the altered history would have "foo(x, y, z)" in E,
>> >> "foo(x,y,z)" in X, and "foo(x,y)" in A.  If you start blaming from F,
>> >> you'd get E as the commit that explains the latest state.  If you do
>> >> not skip Y, you'd get Y.  I am not sure how you'd get X in either
>> >> case.
>> >
>> ...
>
> From my own community, this came up also. The intent was to show everyone
> who touched a particular line, throughout history, not just the current one.
> Perhaps that is what Barret is going for.

I think I now understand what is going on. In short, what Barret wanted is way more ambitious than "blame in a hypothetical history that lacks certain commits". There is no similiarity with existing "-S <revs>" option that I asked about in my initial review. This is quite a different beast (and I kind-a like it ;-).

Instead of "skipping" Y, Barret wants to pretend that the effect of commit Y, in addition to the commit Y itself, never existed in the history. Even when blaming back from F (where it has "foo(x, y, z)", which is the same as E and down to Y), the algorithm wants to pretend as if E and F had that line in "foo(x,y,z)" form, i.e. before Y touched it, and also pretend that the user started blaming a line in that shape.

As there is no general solution [*1*] to "how did this line looked like before this commit", such a blame may not have a meaningful answer. When Y added a block of text without removing anything (i.e. "git show Y" has no line that begins with '-'), and when the user asked to blame that block of text in F, there is no sensible answer to the question: "what got replaced by Y in this text?", so a blame that "ignores" Y would not be able to answer "where did these lines come from?" question. But in some cases (e.g. when "git show Y" has only one '-' line, immediately followed by '+' line, both with quite a similar contents and no other change in the patch---it is highly plausible that in such a change, the former became the latter, like the example in Barret's message I was responding to), we can say "this was how the line that was modified in Y looked like before Y" and blaming while ignoring Y can have a reasonable answer.

[Footnote]

*1* Here, I use the word "solution" in the same sense as "solution to a math equation", not in the sense to describe the steps to derive that "solution".